NWI-I00041 Information Retrieval Index

Course organization: Faegheh Hasibi.

  • Assignment sessions on friday start in October.
  • Required readings before class
  • Grading: 50/50 assignment & end term exam; both need to be passed.

Exam material:

  • Slides
  • Lecture material
  • Key points from mandatory background reading

Assignment:

  • Carry out an IR experiment and write a report about it
    • Proposal included
    • Group work, 2-3 people

Links:

  • tiny.cc/register-IR2021
  • tiny.cc/join-slack-IR2021
  • facebook.com/informagus
  • twitter.com/informagus
  • youtube channel informagus

Programme

  1. Introduction
  2. Information Retrieval Models
  3. Retrieval Models continued
  4. Implementation: Indexing and Inverted Files
    • TODO SEIRIP Ch. 5 (quite detailed; hope the lectures are enough)
  5. Evaluation: Cranfield Paradigm
  6. Entity retrieval and linking
    • EOS book Ch. 3 and Ch. 5; incorporated in lecture notes
    • optional: 3.3.1.3, 3.3.2.2., 3.3.2.4, 3.3.3, 3.5.2, 5.6.2-5.7, 5.8.2-6
  7. Relevance feedback
    • IIR ch. 9.1, about Roccio; sufficiently explained in the lecture notes
    • SEIRiP ch7.3.2
  8. Learning to rank
  9. Federated Search
  10. Spoken content retrieval
  11. Link analysis
  12. Neural Networks for IR
    • Focus on weak supervision; reading material is also about that.
  13. Conversational Search
  14. Learning to rank from user interactions

Exam prep

Observations from reviewing older exams.

  • Many questions about interpreting formula components
    • "Why would the researchers who proposed this scheme use the logarithm of the term frequency?"
    • "What is the effect of the exponential function in the formula?"
  • Open-ended questions where reasoning is expected
    • "Explain how a future conversational search system may provide new ways for a system to discover what makes a result relevant."
  • Connecting and contrasting fields
    • "What aspect of information retrieval makes it less amenable to the application of machine learning?"
  • Questions about implementation
    • "How do we use the two indexes during query processing? Assume the system uses BM25 as its retrieval model. Where are the quantities that we need to compute document scores stored in these data structures?"
  • Evaluation schemes and their limitations
    • "Based on what information does counterfactual evaluation estimate the relevance information needed to evaluate a retrieval system?"
    • Explain them; write their formulas
    • When to use which evaluation measure and/or paradigm
  • Questions that ask for diagrams/illustrations:
    • "Explain in detail how a pointwise learning to rank approach, e.g. using logistic regression, would re-rank the documents. Draw a diagram that illustrates the learning to rank setup."
    • Or how indexes work, e.g. with BM25
  • Questions about experimental designs in IR
    • "How would you evaluate the impact of ranking by a combination of readability and topicality on the search engine effectiveness?"
  • Specific knowledge questions about one of the topics (e.g. semantic search, the web/PageRank, Federated search algorithms etc.)

See the IR Quiz .

Concepts